OpenLoaf

Triggered when the user asks the AI to directly operate their macOS desktop: open an app, click a button, fill a form inside a native window, read what's currently on screen, automate a local GUI flow. Typical phrasing: "in Finder, please…", "click X in the Y app", "what's on my screen right now". **Available only in OpenLoaf Desktop (macOS)**; other platforms return a desktop-only error — fall back to `BrowserAct` or hand the task back to the user. **Not for**: in-page web interaction (→ `browser-ops-skill`), local file I/O (→ `Read`/`Edit`/`Write`).

当用户要求 AI 直接操作他的 macOS 桌面时触发：打开某个 App、点击某个按钮、填写窗口里的表单、读取当前屏幕内容、自动化本地 GUI 流程。典型说法"帮我在访达里…"、"帮我在某 App 里点一下…"、"看看我屏幕上显示的是什么"。**仅在 OpenLoaf Desktop（macOS）中可用**；非桌面端会返回 desktop-only 错误，应改用 `BrowserAct` 或让用户自己操作。**不用于**：网页内交互（→ `browser-ops-skill`）、本地文件读写（→ `Read`/`Edit`/`Write`）。

Trigger whenever the user wants to do anything with a real email account: check inbox, read messages, reply, forward, compose, send, search, archive, delete. Typical phrasing: "check my inbox", "what did X send me", "查下我邮箱". **Do not use for**: instant messaging / IM / WeChat / Slack, translating text that merely contains the word "email", or offline parsing of `.eml` files (→ just use Read).

当用户要对真实电子邮件账户做任何操作时触发：查收件箱、读信、回复、转发、撰写、发送、搜索、归档、删除。典型说法"查下我邮箱"、"xx 给我发了啥"、"check my inbox"。**不用于**：即时消息 / IM / 微信 / Slack、翻译含"邮件"二字的文本、对 .eml 文件做离线解析（→直接用 Read）。

**Required for visualized output.** When the prompt involves news/briefings/market updates/rankings/reports/summaries/analyses/comparisons/selection/recommendations/roundups/retrospectives, or is expected to return 3+ structurable items, **this skill must be loaded before (or in the same turn as) the data-fetching tool**. Once loaded, **do not finish with plain markdown text**—you must render via JsxCreate or ChartRender. Typical examples: "search the latest news about Iran", "compare Obsidian/Notion/Logseq", "recommend a few books", "how is the stock market recently". **Not for**: single-fact Q&A, pure code output, adding/removing nodes inside the canvas (handled by the canvas sub-agent).

**输出可视化，必加载**。当 prompt 涉及新闻/资讯/行情/排行/报告/总结/分析/对比/选型/推荐/盘点/复盘，或预计返回 3 条以上可结构化条目时，**必须先于（或同轮于）数据获取工具加载本 skill**。加载后**禁止用纯 markdown 文本收尾**——必须用 JsxCreate 或 ChartRender 渲染。典型："搜最新伊朗新闻"、"对比 Obsidian/Notion/Logseq"、"推荐几本书"、"最近股市怎么样"。**不用于**：单句事实问答、纯代码输出、在画布里增删节点（由画布子 Agent 处理）。

Triggers whenever the user asks to do anything with PowerPoint slides (.pptx): read, summarize, extract, edit, create, or convert. Key insight: PPTX files are visually structured — charts, diagrams, and layouts cannot be reliably captured by text extraction alone. The correct approach for understanding slide content is to render the target slide(s) as PNG images (via PptxInspect render) and pass them to a vision model (CloudImageUnderstand), not to rely solely on text/outline extraction. Use text tools (outline / text / shapes) only as supplementary context. Typical phrasings: "summarize this PPT", "what is this deck about", "what's on slide 3", "describe page 5", "make me a Q4 report deck", "change the title of slide 3". Load this skill whenever the user mentions a .pptx file, deck, slides, or presentation.

当用户要求对 PowerPoint 幻灯片（.pptx）做任何操作时触发：阅读、总结、提取、编辑、创建、格式转换均适用。核心认知：PPTX 是视觉结构化文件——图表、排版、示意图等内容靠纯文字抽取无法可靠还原。理解幻灯片内容的正确路径是：用 PptxInspect render 把目标页渲染成 PNG，再交给视觉模型（CloudImageUnderstand）理解，而不是仅靠 text/outline 文字提取。文字工具仅作补充上下文使用。典型说法："总结这份 PPT"、"这个 deck 在讲什么"、"第 3 页讲什么"、"描述第 5 页内容"、"帮我做一份 Q4 汇报 PPT"、"改第 3 页标题"。用户提到 .pptx 文件、deck、幻灯片、演示文稿时都加载本技能。

Triggered when the user wants to create, open, switch, move, delete, or rename OpenLoaf's "project" entity. Typical phrasings: "new project", "add ~/code/foo", "list all projects". **Not for**: reading/writing files inside a project (→ use Read/Edit/Write directly), discussing "project planning / requirement docs" (→ answer directly), routine Git operations (→ `Bash`).

当用户要对 OpenLoaf 的"项目"这一实体做创建、打开、切换、移动、删除、改名时触发。典型说法"新建项目"、"把 ~/code/foo 加进来"、"列出所有项目"。**不用于**：项目内文件读写（→直接用 Read/Edit/Write）、讨论"项目规划 / 需求文档"（→直接回答）、纯 Git 日常操作（→`Bash`）。

All-in-one PDF read / write / convert / OCR. Trigger scenarios: summarize a PDF, extract text or tables, fill an AcroForm, render pages as images, merge, split, add a watermark, rotate, encrypt/decrypt, create a new PDF (invoice / report / receipt), convert between PDF and docx/md/txt. Typical phrasings: "summarize this PDF", "fill out this PDF form", "merge these PDFs", "add a watermark to the PDF", "convert PDF to Word", "OCR this scan", "show me what page 3 of the PDF looks like". Load this skill whenever the user mentions a .pdf file, needs to produce a PDF, or wants to make any change to a PDF.

PDF 读/写/转/OCR 一体化。触发场景：总结 PDF、读文本或表格、填 AcroForm 表单、渲染页面为图、合并/拆分/水印/旋转/加密、创建新 PDF（发票/报告）、PDF ↔ docx/md/txt 互转。典型说法：“总结这份 PDF”、“填一下这个 PDF 表单”、“把几个 PDF 合一起”、“PDF 加水印”、“PDF 转 Word”、“扫描件 OCR”、“看看 PDF 第 3 页长什么样”。用户提到 .pdf 文件、要产出 PDF、或要对 PDF 做任何改动都加载本技能。

All-in-one Word document (.docx) read / write / convert / review. Trigger scenarios: summarize a docx, read paragraphs / outline / tables / images / comments / tracked changes, replace body text, insert images, change page settings, rebuild table of contents, add comments or replies, add tracked changes (insert/delete/replace), accept/reject revisions, convert docx ↔ pdf/html/md/txt. Typical phrasings: "summarize this Word doc", "change the second paragraph to XXX", "replace all 'Company A' with 'Company B'", "add a tracked change comment", "accept all revisions", "convert docx to pdf", "generate a sales report from these bullet points". Load this skill whenever the user mentions a .docx / .doc file or targets a Word document as the deliverable.

Word 文档（.docx）读/写/转/评审一体化。触发场景：总结 docx、读段落/大纲/表格/图片/评论/修订、替换正文、插入图片、改页面设置、重建目录、加评论或回复、加修订（insert/delete/replace）、接受/驳回修订、docx ↔ pdf/html/md/txt 互转。典型说法："帮我总结这份 Word"、"第二段改成 XXX"、"把全部'A 公司'换成'B 公司'"、"加一条修订批注"、"接受所有修订"、"docx 转 pdf"、"从这些要点生成销售报告"。用户提到 .docx / .doc 文件或以 Word 文档为产出目标都加载本技能。

Unified read / write / stats / recalc / chart capability for Excel / CSV (.xlsx / .xlsm / .csv / .tsv). Trigger scenarios: reading cells, summarizing, grouping, highest/lowest values, total revenue, editing cell values or formulas, inserting rows/columns, adding total rows, inserting charts (bar/line/pie), freezing panes, auto-filter, conditional formatting, CSV ↔ XLSX conversion, building rosters / sales sheets / financial models, diagnosing #REF! / #DIV/0! formula errors, triggering recalculation. Typical phrasings: "who has the highest sales in this sheet", "total Q1 revenue", "change B2 to 500", "add a total row", "create a new roster", "convert csv to xlsx", "freeze the first row", "highlight negatives in column A red". Load this skill whenever the user mentions an .xlsx / .csv file or the final deliverable is a spreadsheet.

Excel / CSV（.xlsx / .xlsm / .csv / .tsv）读/写/统计/重算/图表一体化。触发场景：阅读单元格、汇总分组、谁最高谁最低、总收入、改单元格或公式、加行列、加合计行、插入图表（bar/line/pie）、冻结首行、加筛选、条件格式、CSV ↔ XLSX 互转、新建花名册/销售表/财务模型、排查 #REF!/#DIV/0! 公式错误、触发重算。典型说法："这表里谁销量最高"、"Q1 总收入"、"把 B2 改成 500"、"加一行合计"、"新建花名册"、"csv 转 xlsx"、"冻结首行"、"A 列负数标红"。用户提到 .xlsx / .csv 文件或最终产出是表格都加载本技能。

当主 Agent 面临多步骤复杂任务并正在判断要不要 / 怎样把子任务外包给内置子代理（browser / doc-editor / data-analyst / extractor / canvas-designer / coder 等）时触发。**不用于**：单步问答 / 读取 / 简单副作用（主 Agent 直接做）、已经明确要用 `Agent` 且知道 subagent_type 的场景。

当用户要求对具体网页做页面级交互时触发：登录、填表、点按钮、翻页抓取、截图、下载页面图片、处理验证码或反爬、访问 SPA 动态内容；也用于 WebFetch 拿到空壳或被风控后的降级。典型说法"帮我登录 X 然后抓 Y"、"截个图"。**不用于**：事实性问题或"xx 是什么"（→`WebSearch`）、单纯读取静态网页文本（先试 `WebFetch`）。

当用户要对真实日历数据做查询或增删改时触发：看今天/本周安排、查冲突、建会议、改时间、删日程。典型说法"我明天几点开会"、"帮我约 xx"、"schedule a meeting"。**不用于**：对话里口头提到时间或倒计时（→直接回答）、周期性后台任务如"每天 9 点跑脚本"（→schedule-ops-skill）。

当用户要对 OpenLoaf 画布 / 白板做生命周期管理时触发：新建、打开、筛选、复制、删除、改标题或归属。典型说法"开个白板"、"创建一个架构图画布"、"清理没用的白板"。**不用于**：数据图表可视化（→visualization-ops-skill）、AI 生成图片或视频（→cloud-media-skill）、在已打开画布里编辑节点（由画布子 Agent 处理）。

Process local image / video / audio files, or download videos from the web. Images: resize, crop, rotate, convert format (→webp/png/jpeg), apply filters, inspect dimensions. Video: convert format, extract audio, inspect metadata. Download: public videos from YouTube / Bilibili etc. Also the go-to skill for post-processing AI-generated images (format conversion, resizing). **Not for**: AI generating images/video/voice from scratch (→cloud-media-skill), or rendering charts inline (→visualization-ops-skill).

处理本地已有的图片 / 视频 / 音频文件，或下载网络视频。图片：缩放、裁剪、旋转、格式转换（→webp/png/jpeg）、滤镜、查看尺寸；视频：格式转换、提取音频、查看元信息；下载：YouTube / B 站等公开视频。也用于 AI 生成图片后的二次处理（转格式、缩放）。**不用于**：AI 从零生成图片/视频/语音（→cloud-media-skill）、在对话里渲染图表（→visualization-ops-skill）。

当用户要把任务交给后台异步执行、按 cron / 周期性重复、委派给项目 Agent 长时间跑，或查看/审批/取消看板任务时触发。典型说法"每天 9 点帮我 X"、"让 coder agent 跑这个"、"有哪些待审批"。**不用于**：一次性即时答复（→直接用工具）、真实日历会议 / 约会（→calendar-ops-skill）、一次性计划审批（→`SubmitPlan`）。

当用户想要创建、编辑、改进或调试 OpenLoaf 自定义技能（Skill）时触发。典型说法："帮我创建一个技能"、"做一个新 skill"、"把刚才的操作封装成技能"、"写一个能自动 XX 的技能"、"改一下这个 skill"、"这个 skill 为什么不触发"、"编辑我的自定义技能"、"加个全局技能"、"给当前项目加个技能"。任何涉及 `.openloaf/skills/` 目录下 `SKILL.md` 的创建 / 修改 / 调优请求都应加载本技能。也适用于用户想理解技能格式、排查触发问题、或把对话里的工作流固化成可复用能力的场景。不用于：内置技能（如 file-ops、email-ops 等）的修改——那些是平台随版本发布的只读能力。

当用户要在 OpenLoaf 工作台 / 桌面上新增、修改、查看、删除 widget（桌面组件 / 小组件 / 仪表盘卡片）时触发。典型说法"加个天气 widget"、"做个番茄钟组件"、"改下这个 widget 的颜色"。**不用于**：聊天消息里一次性展示图表（→visualization-ops-skill）、AI 生成图片（→cloud-media-skill）、对话里随口提到"时钟 / 倒计时"（→直接回答）。

Triggers when the master Agent faces a multi-step complex task and is deciding whether / how to outsource sub-tasks to built-in subagents (browser / doc-editor / data-analyst / extractor / canvas-designer / coder, etc.). **Not for**: single-step Q&A / reads / simple side effects (master Agent handles directly), or scenarios where `Agent` is already confirmed and `subagent_type` is known.

Triggered when the user asks for page-level interaction with a specific webpage: login, form filling, button clicks, pagination scraping, screenshots, downloading page images, handling CAPTCHAs or anti-bot measures, accessing SPA dynamic content; also used as a fallback when WebFetch returns an empty shell or gets blocked. Typical phrasing: "log me into X then scrape Y", "take a screenshot". **Not for**: factual questions or "what is xx" (→ `WebSearch`), or simply reading static webpage text (try `WebFetch` first).

Triggers when the user wants to query or modify real calendar data: checking today's or this week's agenda, spotting conflicts, creating meetings, rescheduling, or deleting events. Typical phrasings include "what time is my meeting tomorrow", "book xx for me", or "schedule a meeting". **Do not use for**: casually mentioning times or countdowns in conversation (→ answer directly), or recurring background jobs like "run this script at 9am every day" (→ schedule-ops-skill).

Triggered when the user wants lifecycle management of OpenLoaf canvases / whiteboards: create, open, filter, duplicate, delete, rename, or change ownership. Typical phrases: "open a whiteboard", "create an architecture diagram canvas", "clean up unused whiteboards". **Not for**: data-driven chart visualization (→visualization-ops-skill), AI image or video generation (→cloud-media-skill), editing nodes inside an already-open canvas (handled by the canvas sub-agent).

Trigger when the user wants to hand a task to background async execution, run it on a cron/recurring schedule, delegate it to a project Agent for a long run, or inspect/approve/cancel tasks on the board. Typical phrasings: "every day at 9am help me X", "let the coder agent run this", "what's pending approval". **Not for**: one-shot immediate answers (→ just use tools directly), real calendar meetings/appointments (→ calendar-ops-skill), one-off plan approvals (→ `SubmitPlan`).

Triggers when the user asks about settings, preferences, API keys, model switching, theme / language / shortcuts / proxy / backup, or related failures (key invalid, model not found) within the OpenLoaf desktop product — where to configure and how to configure them. **Not for**: code configuration files inside a project such as tsconfig / .env (→ use Read/Edit directly), general programming questions (→ answer directly), or system-level OS settings.

当用户问 OpenLoaf 桌面端产品内的设置、偏好、API key、模型切换、主题 / 语言 / 快捷键 / 代理 / 备份、或相关故障（key invalid、model not found）该在哪配置、怎么配置时触发。**不用于**：项目里的代码配置文件如 tsconfig / .env（→直接用 Read/Edit）、一般编程问题（→直接回答）、系统级 OS 设置。

Triggers when the user wants to create, edit, improve, or debug an OpenLoaf custom Skill. Typical phrasings: "help me create a skill", "build a new skill", "turn what we just did into a skill", "write a skill that auto-XX", "tweak this skill", "why isn't this skill triggering", "edit my custom skill", "add a global skill", "add a skill to the current project". Any create / modify / tune request touching a `SKILL.md` under `.openloaf/skills/` should load this skill. Also applies when the user wants to understand the skill format, troubleshoot triggering issues, or solidify a conversation workflow into a reusable capability. Not for: modifying built-in skills (e.g. file-ops, email-ops) — those are read-only capabilities shipped with the platform.

Triggered when the user wants to add, modify, view, or delete widgets (desktop components / mini apps / dashboard cards) on the OpenLoaf workbench / desktop. Typical phrasings: "add a weather widget", "build a pomodoro component", "change the color of this widget". **Not for**: one-off chart rendering inside a chat message (→visualization-ops-skill), AI image generation (→cloud-media-skill), casually mentioning "clock / countdown" in conversation (→answer directly).